Written by an AI agent working in this repository.
On 2026-10-31 the owner set four overarching directions for this repository: deep tech industrial, energy, real estate development and construction, and a construction data warehouse, with robotics, site imaging and sensing, and autonomy inside them. The reason given was where investment is heading. That is the owner's read of the market, not something this repository has measured.
Most of the groundwork exists already. Of the 25 research tracks, nine fit the new directions:
| Track | Projects on main today |
|---|---|
12 house: open residential construction | 40 |
20 buildbot: robots on the building site | 26 |
25 heavy: construction, mining and heavy equipment | 1 |
15 sensing: imaging, sensor fusion, 3D from captures | 31 |
16 energy: supply and dense storage | 27 |
21 infra: the physical industries | 22 |
9 osrobot, 22 aerolog, 23 marine: robots and autonomy | 40, 1, 2 |
So the change is about where effort goes, not about starting from zero. Three changes, each in a file:
- Research draws.
research/_shared/pick_track.pynow sends half of all track draws to these nine tracks only; the other half stays uniform over all 25, so about 68% of track runs and 75% of programme runs land in the focus areas. No track is closed.houseandosrobotare past the 40-project mark at which the picker redraws, so a focus draw that hits them moves to an open focus track. In practice that means heavy equipment, aerial and water autonomy, energy, sensing and site robots. - Products.
wall-framing-takeoff, the framing, cutting-stock and takeoff engine for new homes, is now the first flagship. The bookkeeper is second, not dropped. - The data layer. A new research programme,
build-data, asks whether public building-permit records from several cities, loaded onto one open schema, predict how long permits and builds take better than each city's own averages. It is the first table of the construction data warehouse: real, licence-checked public records that the research and the products can both measure against. It has no proposal yet; the first run that draws it writes one.
What will not change: no claims of partners, traction or funding; designs stay research, and nothing runs on a real machine, site or grid without the owner.